Stochastic kinetics of the circular gene hypothesis: feedback effects and protein fluctuations

0.29 MB·English

by  Raoul R. Wadhwa

Stochastic kinetics of the circular gene hypothesis:
feedback effects and protein fluctuations
5 R.R. Wadhwaa, L. Zala´nyib, J. Szentec, L. N´egyessyb, P. E´rdia,b,∗
1
0
aCenterfor Complex Systems Studies, Kalamazoo College, 1200 Academy Street,
2
Kalamazoo, MI 49006, USA
g bWigner Research Centre for Physics, Hungarian Academy of Sciences, Budapest, Hungary
u cAtmospheric, Oceanic and Space Sciences, College of Engineering, Universityof Michigan,
A Ann Arbor, MI, USA

1
1

]
N Abstract
M Stochastic kinetic models of genetic expression are able to describe pro-
tein fluctuations. A comparative study of the canonical and a feedback model
.
o
is given here by using stochastic simulation methods. The feedback model is
i
b skeleton model implementation of the circular gene hypothesis, which suggests
- the interaction between the synthesis and degradation of mRNA. Qualitative
q
and quantitative changes in the shape and in the numerical characteristics of
[
the stationary distributions suggestthat more combined experimental and the-
2
oreticalstudies shouldbe donetouncoverthe detailsofthe kinetic mechanisms
v
of gene expressions.
9
5
Keywords: genetic expression, stochastic kinetics
3
2010 MSC: 80A30
3
0
.
1
1. Introduction
0
5
1 Protein availability is a condicio sine qua non of cellular processes and sur-
: vival, and is determined by gene regulation. Gene regulation contains many
v
biochemical and biophysicalprocesses. While traditional biochemistry adopted
i
X a rather rigid deterministic scenario considering the execution of instructions
r encodedinDNA,chemicalreactionstakingplaceatthe singlecelllevelarenow
a
admittedly better described by stochastic models than by deterministic ones.
Reactionsingeneexpression,suchaspromoteractivityandinactivity,transcrip-
tion, translation, and decaying of mRNA and proteins are the most important
chemicalsteps. Measurementsonstochastic gene expressionin single cells with
single molecule sensitivity [6, 9] implied the necessity of stochastic description
[22]. Since our goal here is to contribute to the understanding of the nature of
proteinfluctuations,onlystochastickineticmodelingtechniquecanberelevant.
∗Correspondingauthor: [phone]
Email address: [email protected] (P.E´rdi)
Preprint submitted toMATCOM (Elsevier) August 12, 2015
The perspective that models of gene expression should have stochastic ele-
ments goes back to the pioneering works of D. Rigney and O. Berg [3, 23–25],
butthese workscametoo earlyformainstreammolecularbiologists. Stochastic
chemicalkineticsbecamethelinguafranca ofmodelinggeneregulatorynetworks
and related fields twenty years later due to highly cited papers [2, 18].
Stochastic models proved to be very efficient to study the kinetic mecha-
nisms of genetics and, more generally, systems biological processes [10]. Mea-
sured fluctuations in reactions two sources [9] (i) intrinsic noise is related to
variationsin protein levels evenin a populationof cells with identical genotype
and concentrations and states of cellular components, (ii) extrinsic noise due
tofluctuationsinthe amountoractivityofmoleculesinvolvedinthe expression
ofagene,likeRNApolymeraseorribosomes. Thereactionsystem,whatmight
be called the canonical model of gene expression [20], belongs to the category
of compartmental models. Such systems are characterized by the fact that the
activity of one molecular entity is independent of the other entities. In other
words, no interaction between any two such entities occurs. Such models can
fully be solved (i.e. the time-dependent moments can be calculated) by using
the generating function method. More specifically, the different sources of pro-
teinfluctuations havebeencalculated[20]. Under nottoo restrictiveconditions
it was found [6, 11] that the stationary distribution of the protein fluctuation
can be well approximated by gamma distribution. However, as gamma distri-
butionis averygeneralone,alterationsofthe reactionssystemand/orthe rate
constants may imply changes in the shape and parameters of the stationary
distributions.
Realistic models should take into accountfeedback, burst, delay, etc. mech-
anisms too, and some exact results are available for specific families of models
[4, 11, 14, 21, 27].These models contain bimolecular reaction steps too, so the
compartmentalkineticframeworkbasedonindependentactivitiescannotbeas-
sumed anymore. Linear noise approximation is often used to calculate protein
fluctuations,butitsreliabilityforsystemscontainingbimolecularreactionsteps
is restricted [29].
Simulationmethods(forarecentshortreviewsee2.6of[10])areappropriate
tools to obtain information about the size and nature of fluctuations as they
help overcome the limitations of the methods described above.
A recent conceptually new hypothesis [13] suggested that gene expression
mightbecircular,sincethedegradationandsynthesisofmRNAseemtobeinter-
connectedbyafeedbackmechanism. Duetothelackofavailablekineticdatathe
hypothesis cannot be falsified for the time being. However, as the metabolism
ofmRNAisbetterdescribedbysomebimolecularreactions,itmightaffectpro-
tein fluctuations. Specifically, by setting a secondary mechanism to promote
mRNA synthesis may increase the lifetime ratio of the lifetimes of mRNA and
of proteins, which increases protein fluctuation. Our question was whether or
not the feedback mechanism has a significant effect on protein fluctuations. If
yes, it is worth studying the details.
2
2. Biological background
Geneexpressionisthecomplicatedprocessofconvertinggeneticinformation
from a DNA sequence into proteins. In eukaryotes, DNA is located in the cell
nucleus. Prokaryotes do not have a nucleus, and DNA can be found in the
cytoplasm. In prokaryotes there are two main processes in gene expression:
transcription and translation. In eukaryotes, there is an additional process:
splicing.
Transcription is a series of events that use DNA to synthesize messenger
RNA(mRNA)byusingtheenzymeRNApolymeraseasacatalyst. Theseriesof
events contain (in prokaryotes) binding, initiation, RNA synthesis, elongation,
and termination. Specifically, a promoter is a region of DNA where binding
of transcription factor proteins initiate transcription. Eukaryotic transcription
is much more complicated, but depends on these basic steps.
Splicing is a modification of the nascent mRNA transcript in which cer-
tain nucleotide sequences (introns) are removed while other sequences (exons)
remain.
Translation is a process in which the is read out from the mRNA by the
ribosomecomplexandtranslatedintotheaminoacidsequenceinproteins(with
the help of tRNA). It contains more elementary steps, such as initiation, elon-
gation, translocation, and termination.
Degradation: althoughDNAisstable,RNAandproteinmoleculesaresub-
jectto degradation. Itis animportantstepin the regulationofgene expression
and fluctuation of protein concentration.
A recent hypothesis [13] suggested that eukaryotic gene expression can be
viewed as a circular process, where transcription and mRNA degradation are
interconnected. Thebigquestionis,“HowcouldmRNAsynthesisinthenucleus
and mRNA decay in the cytoplasm be mechanistically linked?” [5]. Possible
mechanismsof coupling mRNA synthesis anddecayhave beenanalyzed[5, 19].
′ ′
The 5 to 3 exoribonuclease xrn1, a large protein involved in cytoplasmatic
mRNAdegradationsmightbe acriticalcomponent[19],anditmayplayadual
role in some subprocesses of transcription, namely in initiation and elongation.
Basedontheseobservationsaboutthedualroleofxrn1intranscription[13,
19], a minimal model that takes into account feedback effects has been set by
including three more steps: promoter assignment, promoter reassignment, and
xrn1 dependent transcription.
In this paper, the nature of protein fluctuation in the canonical model and
a simple feedback model implementing the dual role of xrn1 is studied using
stochastic simulations. Based on the results, we predict that the feedback pro-
cess has significant effects on the fluctuations by the additive effects of the
enhancedmRNAfluctuations,sothe detailedmechanismsshouldbe studiedby
combined experimental and modeling studies.
3
3. The model
3.1. Canonical model
Gene expression can be modeled as a three-stage process: gene activation,
transcription, and translation. These are coupled by the opposite processes
of gene inactivation, mRNA degradation, and proteolysis, respectively. Gene
expressioncanbemodeledasareactionsystemofthreechemicalspecies,slightly
modified from the existing schematic for the canonical model [4].
Gene activation: inactive gene −−λ−1→ active gene
Gene inactivation: active gene −−λ−2→ inactive gene
ρ1
Transcription: active gene −−−→ active gene + mRNA
ρ2
mRNA degradation: mRNA −−−→ ∅
γ1
Translation: mRNA −−−→ mRNA + protein
γ2
Proteolysis: protein −−−→ ∅
In this system, all the reactions are first order. This reaction system can be
alternatively defined using the number of each chemical species present in a
cell. Here n , n , n , and n represent the number of inactive genes, active
1 2 3 4
genes, mRNAs, and proteins in the cell, respectively [20].
Gene activation: (n ,n ) −−λ−1−n−1→ (n −1,n +1)
1 2 1 2
Gene inactivation: (n ,n ) −−λ−2−n−2→ (n +1,n −1)
1 2 1 2
ρ1n2
Transcription: n −−−−−→ n +1
3 3
ρ2n3
mRNA degradation: n −−−−−→ n −1
3 3
γ1n3
Translation: n −−−−−→ n +1
4 4
γ2n4
Proteolysis: n −−−−−→ n −1
4 4
To determine the value of the rate constants, we refer to experimental results,
using E. coli as our model organism. The half-life of mRNA in E. coli, cal-
culated as the natural logarithm of 2 divided by the rate constant for mRNA
degradation, is between 3 min and 8 min [4]. Using a timescale measured in
seconds,wechoosethe averagehalf-life ofanmRNA moleculeinthe simulation
to be 300 s. This leads to a value of ln(2)/300≈0.00231 for ρ .
2
Thereareindicationsfromexperimentaldatathatρ /ρ isvariableinE. coli,
1 2
ranging from 1 to a few dozen - we assume a value of 10 for this ratio, which
implies ρ ≈0.0231[4,16]. It hasalsobeen experimentally determinedthat for
1
proteinsinE. coli, theaveragevalueofγ /γ is540[4,16]. Forthis simulation,
1 2
we choose γ = 0.14 as it approaches a stationary state with sufficient speed.
1
For the sakeof simplicity, we assume there exists only a single copy of the gene
we are interested in. We also choose λ = 1 and λ = 7, although we shall
1 2
see that the choice of values for λ and λ is arbitrary. The ratio λ /(λ +λ )
1 2 1 1 2
4
indicatestheproportionoftimeforwhichageneisactive[20],andiswhattruly
matters.
The initial value of (n ,n ,n ,n ) for the reaction system was (1,0,0,0).
1 2 3 4
3.2. Feedback model: the role of xrn1
Gene expression involving xrn1 requires a model with more reactions to be
accuratelymodeled. UsingthebiologicalbackgroundgiveninSection2,wegive
the following model to account for feedback due to xrn1. Here n , n , and n
5 6 7
represent the number of xrn1 molecules, xrn1 complexes, and xrn1 binding to
the promoter in the cell, respectively.
Gene activation: (n ,n ) −−λ−1−n−1→ (n −1,n +1)
1 2 1 2
Gene inactivation: (n ,n ) −−λ−2−n−2→ (n +1,n −1)
1 2 1 2
ρ1n2
Transcription: n −−−−−→ n +1
3 3
Promoter assignment: (n ,n ,n ) −−ω1−n−2−n→6 (n −1,n −1,n +1)
2 6 7 2 6 7
Promoter reassignment: (n ,n ,n ) −−ω−2−n−7→ (n +1,n +1,n −1)
2 5 7 2 5 7
ρ3n7
xrn1 dep. transcription: n −−−−−→ n +1
3 3
γ1n3
Translation: n −−−−−→ n +1
4 4
ρ2n3n5
mRNA degradation: (n ,n ,n ) −−−−−→ (n −1,n −1,n +1)
3 5 6 3 5 6
γ2n4
Proteolysis: n −−−−−→ n −1
4 4
It is experimentally supported that the rate constant of xrn1 dependent
transcription is equal to the rate constant of transcription, implying that ρ =
3
ρ = 0.0231 [13]. The effects of varying values of ω and ω on the resulting
1 1 2
proteindistributionareinvestigatedinSection4. Theremainingrateconstants
inthe feedback model havevalues identical to the correspondingrate constants
in the canonical model.
The initial value of (n ,n ,n ,n ,n ,n ,n ) for the reaction system was
1 2 3 4 5 6 7
(1,0,0,0,10,0,0).
Rate Constant Value
λ 1.0
1
λ 7.0
2
ρ 0.0231
1
ρ ρ /10
2 1
γ 0.14
1
γ γ /540
2 1
Figure 1: The distribution of proteins in the
canonical model at t = 100,000 s with n = 10,000 Table 1: The rate constants
stochastic trajectories. The plotted lineis the best- usedtoproducetheproteindistri-
fit gamma distribution with a shape parameter of bution of the canonical model on
11.59±0.16andarateparameterof0.0172±0.0002. the left. ρ1 has been experimen-
Theexpected value isequal to674.54 andthestan- tallydetermined,ashavetheratios
darddeviationofthedistributionis196.80. ρ1/ρ2 andγ1/γ2 [4,16].
5
3.3. Simulation method
All simulations for this paper were conducted in the Cain interface devel-
oped by SeanMauch [17]. Realizations of the stochastic process were produced
using Gillespie’s direct method [12]. Considering a reaction system as a set of
ordinary differential equations as defined by the laws of mass action kinetics,
numericalintegrationusing the Cash-Karpvariantofthe Runge-Kutta method
was conducted to produce deterministic trajectories [7]. All histograms and
plots were producedusing the ggplot2package in the R programminglanguage
with the multiplot function from the Cookbook for R website [1, 30]. Function
fitdistr( ) in the MASS package from the R programminglanguage was used to
find the best fit parameters for the distributions.
Rate Constant Value
λ 1.0
1
λ 7.0
2
ρ 0.0231
1
ρ ρ /10
2 1
ρ ρ
3 1
ω 1
1
ω 3ρ /2
2 1
γ 0.14
1
Figure 2: The distribution of proteins in the γ2 γ1/540
feedback model at t = 100,000 s with n = 10,000
stochastic trajectories. Plotted lines are the best- Table 2: The rate constants
fit gamma distribution with a shape parameter of used to produce the protein dis-
4.089±0.06andascaleparameterof0.020±0.0003. tributionofthefeedbackmodelon
Theexpectedvalueisequalto203andthestandard the left. ρ1 and ρ3 have been ex-
deviationofthedistributionis124.29. perimentally determined, as have
theratiosρ1/ρ2 andγ1/γ2 [4,16].
4. Simulation Results
Stationaryproteindistributionsforbothmodelswereplottedashistograms,
and fitted with gamma distributions. Figure 1 shows the steady state protein
distribution for the canonical model with the typical parameter values given in
Table 1. Figure 2 shows the steady state distribution for the feedback model
with the typical parameter values given in Table 2. The increased skew in
the steady state protein distribution of the feedback model as compared to the
canonical model is observed in all the steady state distributions explored.
4.1. Varying the rate of gene (in)activation: λ and λ
1 2
AsstatedinPaulsson(2005),thekineticdynamicsofgeneactivationandin-
activationcanbe approximatelymodeledasa randomtelegraphprocess,where
eachgeneindependently switches onwithrateλ andswitchesoffwithrateλ .
1 2
TheprobabilityofagenebeingonisgivenbyP =λ /(λ +λ ). Correspond-
on 1 1 2
ingly, the expected number of proteins in steady state should be proportional
to P . This pattern is confirmed by Figure 3, wherein the expected number of
on
6
1/16 1/8 1/4 1/2
0.01 337.5 675 1350 2700
0.05 337.5 675 1350 2700
0.14 337.5 675 1350 2700
0.50 337.5 675 1350 2700
Table 3: Expectednumberofproteinsasgivenbydeterministicsolutionsofthecanonical
model. The horizontal labels show varying values of λ1/(λ1+λ2). The vertical labels show
varyingvaluesofγ1. ρ1=0.0231;ρ2=ρ1/10;γ2=γ1/540.
1/16 1/8 1/4 1/2
0.01 284.3 ± 41.0 568.2 ± 58.2 1136.8 ± 83.1 2275.1 ± 116.5
0.05 336.8 ± 84.8 675.6 ± 120.5 1350.7 ± 173.9 2702.0 ± 241.1
0.14 338.2 ± 139.6 676.4 ± 194.7 1349.7 ± 274.6 2701.0 ± 387.3
0.50 335.1 ± 223.1 675.9 ± 326.2 1342.3 ± 455.3 2705.2 ± 642.1
Table 4: Expected number of proteins ± standard deviation of the steady state protein
distributionas givenbystochastic simulationsof thecanonical model. Thehorizontal labels
showvaryingvaluesofλ1/(λ1+λ2). Theverticallablesshowvaryingvaluesofγ1. ρ1=0.0231;
ρ2=ρ1/10;γ2=γ1/540.
proteins at steady state for the deterministic and stochastic results is linearly
proportionaltoP . DuetothepresenceoftherelationshipdescribedbyPauls-
on
son and confirmation in the canonical model, we arbitrarily choose the values
λ =1andλ =15whenexploringthemorecomplicatedfeedbackmodel. Results
1 2
for other values of λ and λ can be extrapolated by finding the value of P
1 2 on
for the scenario in question.
4.2. Varying the translation rate: γ
1
Interestingly,thedeterministicsolutionsofthecanonicalandfeedbackmod-
els are independent of γ (Figure 3A). In the stochastic results of the canonical
1
model,thenumberofproteinsatsteadystateisloweratγ =0.01;forvaluesof
1
γ = 0.05, 0.14, 0.50 the stochastic results seem to converge to the determinis-
1
tic solution (Figure 3B). However, as seen in Table 4, despite this convergence,
the standard deviation of the steady state protein distribution increases as γ
1
increases. These results suggest that using a single value for the number of
proteins at steady state as given by the deterministic solution becomes an in-
creasingly worse approximation of the true distribution of proteins in a cell as
the translation rate increases.
The convergence pattern does not hold true of the feedback model. Fig-
ure4showsaninstancewherethe deterministicandstochasticsolutionsclearly
divergeirrespective of the value of γ . Thus, for the feedback model, the deter-
1
ministic andstochasticmethods givedistinct valuesforthe expectednumberof
proteins at steady state.
7
Figure 3: The x-axis on both plots is λ1/(λ1+λ2); the y-axis on both plots is the mean
numberofproteinsatsteadystate. (A)Deterministicsolutionsforthecanonicalmodel. The
plottedseriesallmatchexactly(Table3). (B)Stochasticsimulationresultsforthecanonical
model. Theseriesforγ1 =0.05,0.14,and0.50closelyoverlap(Table4).
4.3. Varying the rate of promoter assignment: ω
1
Basedonchemicalreactionrepresentingpromoterassignment,increasingthe
value of ω decreases the number of active genes by 1, decreases the number of
1
xrn1 complexes by 1, and increasing the number of xrn1 binding to promoters
by 1. Decreasing the number of active genes results in reduced occurrence
of transcription; decreasing the number of xrn1 complexes reduces promoter
assignment in a feedback loop; increasing xrn1 binding to promoters increases
xrn1-basedtranscription. Thus,the firsteffect (whichdoesnotdirectly involve
xrn1) should decrease the number of proteins at steady state, while the last
effect (which directly involves xrn1) should increase the number of proteins
at steady state. Overall, as ω increases, the expected number of proteins at
1
steadystatedecreases. This indicatesthatchangingthe numberofactivegenes
byafixedamounthasagreatereffectonthe numberofproteinsatsteadystate
than does changing any of the xrn1 elements by the same fixed amount. In
essence, although adding xrn1-based processes to the canonical model results
in an effect, the basic processes present in both models still have the majority
effect in determining steady state protein levels. Most of the change in protein
levels happens between ω = 0.10 and 0.50.
1
In addition to affecting the mean of steady state protein distribution, in-
creasing ω also results in an overall decrease in the standard deviation of the
1
steady state protein distribution. However, for larger values of γ (e.g. γ =
1 1
0.50,Table 9) the standard deviation is either approximately equal to or larger
than than the expected number of proteins at steady state. This indicates a
highdegreeofvariationinthesteadystatedistribution. Therefore,eventhough
exploringtheeffectofvaryingω inthedeterministicsolution(Figure5E)shows
1
the same trend as it does in the stochastic results, due to the relatively large
values of the standarddeviation, failing to account for variability in the data is
a significant con of the deterministic solution.
8
1.25ρ 1.50ρ 1.75ρ 2.00ρ
1 1 1 1
0.1 156.2 97.9 77.5 67.0
0.5 141.1 92.3 74.1 64.5
1.0 139.4 91.7 73.7 64.2
1.5 138.8 91.4 73.5 64.1
Table 5: Expected number ofproteins as givenbydeterministicsolutions ofthefeedback
model. The horizontal labels show varying values of ω2. The vertical labels show varying
valuesofω1. Muchlikethecanonicalmodel(Table3),deterministicsolutionsofthefeedback
modelareindependentofthevalueofγ1. γ1=0.01,0.05,0.14,and0.50resultedintheexact
samevaluesforexpected numberofproteins.
Figure 4: Deterministic and stochastic solutions for number of proteins at steady state
with the given parameter values: λ1 =1; λ2 = 15; ω1 = 0.01; ω2 = 1.25ρ1. The x-axis on
bothplotsisγ1;they-axisonbothplotsisthemeannumberofproteinsatsteadystate.
1.25ρ 1.50ρ 1.75ρ 2.00ρ
1 1 1 1
0.1 210.3 ± 230.9 78.3 ± 40.8 53.3 ± 18.2 44.0 ± 12.8
0.5 155.2 ± 135.0 68.0 ± 27.9 49.5 ± 15.3 41.5 ± 11.6
1.0 150.2 ± 131.4 66.8 ± 26.7 48.6 ± 14.7 41.0 ± 11.5
1.5 147.8 ± 125.9 65.8 ± 25.3 48.5 ± 14.5 40.9 ± 11.2
Table 6: Expected number of proteins ± standard deviation of the steady state protein
distribution as given by stochastic simulations of the feedback model. The horizontal labels
showvaryingvaluesofω2. Theverticallabelsshowvaryingvaluesofω1. λ1/(λ1+λ2)=1/15;
ρ1=0.0231;ρ2=ρ1/10;ρ3=ρ1;γ1=0.01;γ2=γ1/540.
9
1.25ρ 1.50ρ 1.75ρ 2.00ρ
1 1 1 1
0.1 359.9 ± 482.6 128.0 ± 87.9 88.3 ± 40.5 72.4 ± 27.3
0.5 261.6 ± 307.5 112.8 ± 62.0 81.9 ± 35.1 68.6 ± 24.6
1.0 249.5 ± 288.7 110.3 ± 63.0 79.9 ± 32.4 67.5 ± 23.9
1.5 248.8 ± 310.0 108.7 ± 58.3 80.1 ± 32.4 67.3 ± 23.7
Table 7: Expected number of proteins ± standard deviation of the steady state protein
distribution as given by stochastic simulations of the feedback model. The horizontal labels
showvaryingvaluesofω2. Theverticallabelsshowvaryingvaluesofω1. λ1/(λ1+λ2)=1/15;
ρ1=0.0231;ρ2=ρ1/10;ρ3=ρ1;γ1=0.05;γ2=γ1/540.
Figure 5: Deterministic (E) and stochastic (A-D) solutions for number of proteins at
steadystatewiththegivenparametervalues: λ1 =1;λ2 =15;ρ1 =0.0231. Thex-axisonall
plots isω1; the y-axis onallplots isthe mean number ofproteins at steady state. (A)γ1 =
0.01;(B)γ1 =0.05;(C)γ1 =0.14;(D)γ1 =0.50;(E)Deterministicsolutionisindependent
ofthevalueofγ1.
1.25ρ 1.50ρ 1.75ρ 2.00ρ
1 1 1 1
0.1 376.6 ± 741.3 129.4 ± 137.1 89.3 ± 64.5 72.8 ± 44.1
0.5 265.7 ± 449.3 116.5 ± 107.2 82.2 ± 52.8 69.0 ± 38.4
1.0 247.7 ± 398.9 111.9 ± 98.8 81.3 ± 52.2 68.2 ± 36.9
1.5 249.0 ± 407.0 111.6 ± 98.9 80.4 ± 49.7 68.0 ± 38.9
Table 8: Expected number of proteins ± standard deviation of the steady state protein
distribution as given by stochastic simulations of the feedback model. The horizontal labels
showvaryingvaluesofω2. Theverticallabelsshowvaryingvaluesofω1. λ1/(λ1+λ2)=1/15;
ρ1=0.0231;ρ2=ρ1/10;ρ3=ρ1;γ1=0.14;γ2=γ1/540.
10